Urban Parking Scheme in Hangzhou Based on Reinforcement Learning

Abstract

With the increasing number of motor vehicles in China, the traditional parking schemes have become inefficient. Traditional parking mainly relies on the driver’s experience and judgment, which makes the search for a parking space difficult; at the same time it will cause traffic congestion nearby. The current parking system in Hangzhou, which includes smart payment, smart service, smart supervision and big data analysis, has largely alleviated the aforementioned difficulties. However, the existing smart parking system still has many shortcomings, such as insufficient coverage of hardware, and the software used is only an information display interface, which cannot make any judgements. This paper discusses a parking scheme based on reinforcement learning, including the application of Q-learning and DQN, in order to improve the performance of the parking system. Q-learning is the most likely method for smart parking, and DQN can also be used to improve real-time judgment. This paper compares traditional parking, smart parking, and smart parking with reinforcement learning, and lists their advantages and disadvantages respectively. The comparison shows that smart parking systems are overall more beneficial than traditional systems.